[PATCH] restore sane ->umount_begin() API
[linux-2.6/openmoko-kernel/knife-kernel.git] / fs / namespace.c
blobf48f98110c30aefaa114ff0f15c0b7cece3f938d
1 /*
2 * linux/fs/namespace.c
4 * (C) Copyright Al Viro 2000, 2001
5 * Released under GPL v2.
7 * Based on code from fs/super.c, copyright Linus Torvalds and others.
8 * Heavily rewritten.
9 */
11 #include <linux/syscalls.h>
12 #include <linux/slab.h>
13 #include <linux/sched.h>
14 #include <linux/smp_lock.h>
15 #include <linux/init.h>
16 #include <linux/kernel.h>
17 #include <linux/quotaops.h>
18 #include <linux/acct.h>
19 #include <linux/capability.h>
20 #include <linux/cpumask.h>
21 #include <linux/module.h>
22 #include <linux/sysfs.h>
23 #include <linux/seq_file.h>
24 #include <linux/mnt_namespace.h>
25 #include <linux/namei.h>
26 #include <linux/security.h>
27 #include <linux/mount.h>
28 #include <linux/ramfs.h>
29 #include <linux/log2.h>
30 #include <linux/idr.h>
31 #include <asm/uaccess.h>
32 #include <asm/unistd.h>
33 #include "pnode.h"
34 #include "internal.h"
36 #define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head))
37 #define HASH_SIZE (1UL << HASH_SHIFT)
39 /* spinlock for vfsmount related operations, inplace of dcache_lock */
40 __cacheline_aligned_in_smp DEFINE_SPINLOCK(vfsmount_lock);
42 static int event;
43 static DEFINE_IDA(mnt_id_ida);
44 static DEFINE_IDA(mnt_group_ida);
46 static struct list_head *mount_hashtable __read_mostly;
47 static struct kmem_cache *mnt_cache __read_mostly;
48 static struct rw_semaphore namespace_sem;
50 /* /sys/fs */
51 struct kobject *fs_kobj;
52 EXPORT_SYMBOL_GPL(fs_kobj);
54 static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry)
56 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
57 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
58 tmp = tmp + (tmp >> HASH_SHIFT);
59 return tmp & (HASH_SIZE - 1);
62 #define MNT_WRITER_UNDERFLOW_LIMIT -(1<<16)
64 /* allocation is serialized by namespace_sem */
65 static int mnt_alloc_id(struct vfsmount *mnt)
67 int res;
69 retry:
70 ida_pre_get(&mnt_id_ida, GFP_KERNEL);
71 spin_lock(&vfsmount_lock);
72 res = ida_get_new(&mnt_id_ida, &mnt->mnt_id);
73 spin_unlock(&vfsmount_lock);
74 if (res == -EAGAIN)
75 goto retry;
77 return res;
80 static void mnt_free_id(struct vfsmount *mnt)
82 spin_lock(&vfsmount_lock);
83 ida_remove(&mnt_id_ida, mnt->mnt_id);
84 spin_unlock(&vfsmount_lock);
88 * Allocate a new peer group ID
90 * mnt_group_ida is protected by namespace_sem
92 static int mnt_alloc_group_id(struct vfsmount *mnt)
94 if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL))
95 return -ENOMEM;
97 return ida_get_new_above(&mnt_group_ida, 1, &mnt->mnt_group_id);
101 * Release a peer group ID
103 void mnt_release_group_id(struct vfsmount *mnt)
105 ida_remove(&mnt_group_ida, mnt->mnt_group_id);
106 mnt->mnt_group_id = 0;
109 struct vfsmount *alloc_vfsmnt(const char *name)
111 struct vfsmount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
112 if (mnt) {
113 int err;
115 err = mnt_alloc_id(mnt);
116 if (err) {
117 kmem_cache_free(mnt_cache, mnt);
118 return NULL;
121 atomic_set(&mnt->mnt_count, 1);
122 INIT_LIST_HEAD(&mnt->mnt_hash);
123 INIT_LIST_HEAD(&mnt->mnt_child);
124 INIT_LIST_HEAD(&mnt->mnt_mounts);
125 INIT_LIST_HEAD(&mnt->mnt_list);
126 INIT_LIST_HEAD(&mnt->mnt_expire);
127 INIT_LIST_HEAD(&mnt->mnt_share);
128 INIT_LIST_HEAD(&mnt->mnt_slave_list);
129 INIT_LIST_HEAD(&mnt->mnt_slave);
130 atomic_set(&mnt->__mnt_writers, 0);
131 if (name) {
132 int size = strlen(name) + 1;
133 char *newname = kmalloc(size, GFP_KERNEL);
134 if (newname) {
135 memcpy(newname, name, size);
136 mnt->mnt_devname = newname;
140 return mnt;
144 * Most r/o checks on a fs are for operations that take
145 * discrete amounts of time, like a write() or unlink().
146 * We must keep track of when those operations start
147 * (for permission checks) and when they end, so that
148 * we can determine when writes are able to occur to
149 * a filesystem.
152 * __mnt_is_readonly: check whether a mount is read-only
153 * @mnt: the mount to check for its write status
155 * This shouldn't be used directly ouside of the VFS.
156 * It does not guarantee that the filesystem will stay
157 * r/w, just that it is right *now*. This can not and
158 * should not be used in place of IS_RDONLY(inode).
159 * mnt_want/drop_write() will _keep_ the filesystem
160 * r/w.
162 int __mnt_is_readonly(struct vfsmount *mnt)
164 if (mnt->mnt_flags & MNT_READONLY)
165 return 1;
166 if (mnt->mnt_sb->s_flags & MS_RDONLY)
167 return 1;
168 return 0;
170 EXPORT_SYMBOL_GPL(__mnt_is_readonly);
172 struct mnt_writer {
174 * If holding multiple instances of this lock, they
175 * must be ordered by cpu number.
177 spinlock_t lock;
178 struct lock_class_key lock_class; /* compiles out with !lockdep */
179 unsigned long count;
180 struct vfsmount *mnt;
181 } ____cacheline_aligned_in_smp;
182 static DEFINE_PER_CPU(struct mnt_writer, mnt_writers);
184 static int __init init_mnt_writers(void)
186 int cpu;
187 for_each_possible_cpu(cpu) {
188 struct mnt_writer *writer = &per_cpu(mnt_writers, cpu);
189 spin_lock_init(&writer->lock);
190 lockdep_set_class(&writer->lock, &writer->lock_class);
191 writer->count = 0;
193 return 0;
195 fs_initcall(init_mnt_writers);
197 static void unlock_mnt_writers(void)
199 int cpu;
200 struct mnt_writer *cpu_writer;
202 for_each_possible_cpu(cpu) {
203 cpu_writer = &per_cpu(mnt_writers, cpu);
204 spin_unlock(&cpu_writer->lock);
208 static inline void __clear_mnt_count(struct mnt_writer *cpu_writer)
210 if (!cpu_writer->mnt)
211 return;
213 * This is in case anyone ever leaves an invalid,
214 * old ->mnt and a count of 0.
216 if (!cpu_writer->count)
217 return;
218 atomic_add(cpu_writer->count, &cpu_writer->mnt->__mnt_writers);
219 cpu_writer->count = 0;
222 * must hold cpu_writer->lock
224 static inline void use_cpu_writer_for_mount(struct mnt_writer *cpu_writer,
225 struct vfsmount *mnt)
227 if (cpu_writer->mnt == mnt)
228 return;
229 __clear_mnt_count(cpu_writer);
230 cpu_writer->mnt = mnt;
234 * Most r/o checks on a fs are for operations that take
235 * discrete amounts of time, like a write() or unlink().
236 * We must keep track of when those operations start
237 * (for permission checks) and when they end, so that
238 * we can determine when writes are able to occur to
239 * a filesystem.
242 * mnt_want_write - get write access to a mount
243 * @mnt: the mount on which to take a write
245 * This tells the low-level filesystem that a write is
246 * about to be performed to it, and makes sure that
247 * writes are allowed before returning success. When
248 * the write operation is finished, mnt_drop_write()
249 * must be called. This is effectively a refcount.
251 int mnt_want_write(struct vfsmount *mnt)
253 int ret = 0;
254 struct mnt_writer *cpu_writer;
256 cpu_writer = &get_cpu_var(mnt_writers);
257 spin_lock(&cpu_writer->lock);
258 if (__mnt_is_readonly(mnt)) {
259 ret = -EROFS;
260 goto out;
262 use_cpu_writer_for_mount(cpu_writer, mnt);
263 cpu_writer->count++;
264 out:
265 spin_unlock(&cpu_writer->lock);
266 put_cpu_var(mnt_writers);
267 return ret;
269 EXPORT_SYMBOL_GPL(mnt_want_write);
271 static void lock_mnt_writers(void)
273 int cpu;
274 struct mnt_writer *cpu_writer;
276 for_each_possible_cpu(cpu) {
277 cpu_writer = &per_cpu(mnt_writers, cpu);
278 spin_lock(&cpu_writer->lock);
279 __clear_mnt_count(cpu_writer);
280 cpu_writer->mnt = NULL;
285 * These per-cpu write counts are not guaranteed to have
286 * matched increments and decrements on any given cpu.
287 * A file open()ed for write on one cpu and close()d on
288 * another cpu will imbalance this count. Make sure it
289 * does not get too far out of whack.
291 static void handle_write_count_underflow(struct vfsmount *mnt)
293 if (atomic_read(&mnt->__mnt_writers) >=
294 MNT_WRITER_UNDERFLOW_LIMIT)
295 return;
297 * It isn't necessary to hold all of the locks
298 * at the same time, but doing it this way makes
299 * us share a lot more code.
301 lock_mnt_writers();
303 * vfsmount_lock is for mnt_flags.
305 spin_lock(&vfsmount_lock);
307 * If coalescing the per-cpu writer counts did not
308 * get us back to a positive writer count, we have
309 * a bug.
311 if ((atomic_read(&mnt->__mnt_writers) < 0) &&
312 !(mnt->mnt_flags & MNT_IMBALANCED_WRITE_COUNT)) {
313 printk(KERN_DEBUG "leak detected on mount(%p) writers "
314 "count: %d\n",
315 mnt, atomic_read(&mnt->__mnt_writers));
316 WARN_ON(1);
317 /* use the flag to keep the dmesg spam down */
318 mnt->mnt_flags |= MNT_IMBALANCED_WRITE_COUNT;
320 spin_unlock(&vfsmount_lock);
321 unlock_mnt_writers();
325 * mnt_drop_write - give up write access to a mount
326 * @mnt: the mount on which to give up write access
328 * Tells the low-level filesystem that we are done
329 * performing writes to it. Must be matched with
330 * mnt_want_write() call above.
332 void mnt_drop_write(struct vfsmount *mnt)
334 int must_check_underflow = 0;
335 struct mnt_writer *cpu_writer;
337 cpu_writer = &get_cpu_var(mnt_writers);
338 spin_lock(&cpu_writer->lock);
340 use_cpu_writer_for_mount(cpu_writer, mnt);
341 if (cpu_writer->count > 0) {
342 cpu_writer->count--;
343 } else {
344 must_check_underflow = 1;
345 atomic_dec(&mnt->__mnt_writers);
348 spin_unlock(&cpu_writer->lock);
350 * Logically, we could call this each time,
351 * but the __mnt_writers cacheline tends to
352 * be cold, and makes this expensive.
354 if (must_check_underflow)
355 handle_write_count_underflow(mnt);
357 * This could be done right after the spinlock
358 * is taken because the spinlock keeps us on
359 * the cpu, and disables preemption. However,
360 * putting it here bounds the amount that
361 * __mnt_writers can underflow. Without it,
362 * we could theoretically wrap __mnt_writers.
364 put_cpu_var(mnt_writers);
366 EXPORT_SYMBOL_GPL(mnt_drop_write);
368 static int mnt_make_readonly(struct vfsmount *mnt)
370 int ret = 0;
372 lock_mnt_writers();
374 * With all the locks held, this value is stable
376 if (atomic_read(&mnt->__mnt_writers) > 0) {
377 ret = -EBUSY;
378 goto out;
381 * nobody can do a successful mnt_want_write() with all
382 * of the counts in MNT_DENIED_WRITE and the locks held.
384 spin_lock(&vfsmount_lock);
385 if (!ret)
386 mnt->mnt_flags |= MNT_READONLY;
387 spin_unlock(&vfsmount_lock);
388 out:
389 unlock_mnt_writers();
390 return ret;
393 static void __mnt_unmake_readonly(struct vfsmount *mnt)
395 spin_lock(&vfsmount_lock);
396 mnt->mnt_flags &= ~MNT_READONLY;
397 spin_unlock(&vfsmount_lock);
400 int simple_set_mnt(struct vfsmount *mnt, struct super_block *sb)
402 mnt->mnt_sb = sb;
403 mnt->mnt_root = dget(sb->s_root);
404 return 0;
407 EXPORT_SYMBOL(simple_set_mnt);
409 void free_vfsmnt(struct vfsmount *mnt)
411 kfree(mnt->mnt_devname);
412 mnt_free_id(mnt);
413 kmem_cache_free(mnt_cache, mnt);
417 * find the first or last mount at @dentry on vfsmount @mnt depending on
418 * @dir. If @dir is set return the first mount else return the last mount.
420 struct vfsmount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry,
421 int dir)
423 struct list_head *head = mount_hashtable + hash(mnt, dentry);
424 struct list_head *tmp = head;
425 struct vfsmount *p, *found = NULL;
427 for (;;) {
428 tmp = dir ? tmp->next : tmp->prev;
429 p = NULL;
430 if (tmp == head)
431 break;
432 p = list_entry(tmp, struct vfsmount, mnt_hash);
433 if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) {
434 found = p;
435 break;
438 return found;
442 * lookup_mnt increments the ref count before returning
443 * the vfsmount struct.
445 struct vfsmount *lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
447 struct vfsmount *child_mnt;
448 spin_lock(&vfsmount_lock);
449 if ((child_mnt = __lookup_mnt(mnt, dentry, 1)))
450 mntget(child_mnt);
451 spin_unlock(&vfsmount_lock);
452 return child_mnt;
455 static inline int check_mnt(struct vfsmount *mnt)
457 return mnt->mnt_ns == current->nsproxy->mnt_ns;
460 static void touch_mnt_namespace(struct mnt_namespace *ns)
462 if (ns) {
463 ns->event = ++event;
464 wake_up_interruptible(&ns->poll);
468 static void __touch_mnt_namespace(struct mnt_namespace *ns)
470 if (ns && ns->event != event) {
471 ns->event = event;
472 wake_up_interruptible(&ns->poll);
476 static void detach_mnt(struct vfsmount *mnt, struct path *old_path)
478 old_path->dentry = mnt->mnt_mountpoint;
479 old_path->mnt = mnt->mnt_parent;
480 mnt->mnt_parent = mnt;
481 mnt->mnt_mountpoint = mnt->mnt_root;
482 list_del_init(&mnt->mnt_child);
483 list_del_init(&mnt->mnt_hash);
484 old_path->dentry->d_mounted--;
487 void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry,
488 struct vfsmount *child_mnt)
490 child_mnt->mnt_parent = mntget(mnt);
491 child_mnt->mnt_mountpoint = dget(dentry);
492 dentry->d_mounted++;
495 static void attach_mnt(struct vfsmount *mnt, struct path *path)
497 mnt_set_mountpoint(path->mnt, path->dentry, mnt);
498 list_add_tail(&mnt->mnt_hash, mount_hashtable +
499 hash(path->mnt, path->dentry));
500 list_add_tail(&mnt->mnt_child, &path->mnt->mnt_mounts);
504 * the caller must hold vfsmount_lock
506 static void commit_tree(struct vfsmount *mnt)
508 struct vfsmount *parent = mnt->mnt_parent;
509 struct vfsmount *m;
510 LIST_HEAD(head);
511 struct mnt_namespace *n = parent->mnt_ns;
513 BUG_ON(parent == mnt);
515 list_add_tail(&head, &mnt->mnt_list);
516 list_for_each_entry(m, &head, mnt_list)
517 m->mnt_ns = n;
518 list_splice(&head, n->list.prev);
520 list_add_tail(&mnt->mnt_hash, mount_hashtable +
521 hash(parent, mnt->mnt_mountpoint));
522 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
523 touch_mnt_namespace(n);
526 static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root)
528 struct list_head *next = p->mnt_mounts.next;
529 if (next == &p->mnt_mounts) {
530 while (1) {
531 if (p == root)
532 return NULL;
533 next = p->mnt_child.next;
534 if (next != &p->mnt_parent->mnt_mounts)
535 break;
536 p = p->mnt_parent;
539 return list_entry(next, struct vfsmount, mnt_child);
542 static struct vfsmount *skip_mnt_tree(struct vfsmount *p)
544 struct list_head *prev = p->mnt_mounts.prev;
545 while (prev != &p->mnt_mounts) {
546 p = list_entry(prev, struct vfsmount, mnt_child);
547 prev = p->mnt_mounts.prev;
549 return p;
552 static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root,
553 int flag)
555 struct super_block *sb = old->mnt_sb;
556 struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname);
558 if (mnt) {
559 if (flag & (CL_SLAVE | CL_PRIVATE))
560 mnt->mnt_group_id = 0; /* not a peer of original */
561 else
562 mnt->mnt_group_id = old->mnt_group_id;
564 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
565 int err = mnt_alloc_group_id(mnt);
566 if (err)
567 goto out_free;
570 mnt->mnt_flags = old->mnt_flags;
571 atomic_inc(&sb->s_active);
572 mnt->mnt_sb = sb;
573 mnt->mnt_root = dget(root);
574 mnt->mnt_mountpoint = mnt->mnt_root;
575 mnt->mnt_parent = mnt;
577 if (flag & CL_SLAVE) {
578 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
579 mnt->mnt_master = old;
580 CLEAR_MNT_SHARED(mnt);
581 } else if (!(flag & CL_PRIVATE)) {
582 if ((flag & CL_PROPAGATION) || IS_MNT_SHARED(old))
583 list_add(&mnt->mnt_share, &old->mnt_share);
584 if (IS_MNT_SLAVE(old))
585 list_add(&mnt->mnt_slave, &old->mnt_slave);
586 mnt->mnt_master = old->mnt_master;
588 if (flag & CL_MAKE_SHARED)
589 set_mnt_shared(mnt);
591 /* stick the duplicate mount on the same expiry list
592 * as the original if that was on one */
593 if (flag & CL_EXPIRE) {
594 if (!list_empty(&old->mnt_expire))
595 list_add(&mnt->mnt_expire, &old->mnt_expire);
598 return mnt;
600 out_free:
601 free_vfsmnt(mnt);
602 return NULL;
605 static inline void __mntput(struct vfsmount *mnt)
607 int cpu;
608 struct super_block *sb = mnt->mnt_sb;
610 * We don't have to hold all of the locks at the
611 * same time here because we know that we're the
612 * last reference to mnt and that no new writers
613 * can come in.
615 for_each_possible_cpu(cpu) {
616 struct mnt_writer *cpu_writer = &per_cpu(mnt_writers, cpu);
617 if (cpu_writer->mnt != mnt)
618 continue;
619 spin_lock(&cpu_writer->lock);
620 atomic_add(cpu_writer->count, &mnt->__mnt_writers);
621 cpu_writer->count = 0;
623 * Might as well do this so that no one
624 * ever sees the pointer and expects
625 * it to be valid.
627 cpu_writer->mnt = NULL;
628 spin_unlock(&cpu_writer->lock);
631 * This probably indicates that somebody messed
632 * up a mnt_want/drop_write() pair. If this
633 * happens, the filesystem was probably unable
634 * to make r/w->r/o transitions.
636 WARN_ON(atomic_read(&mnt->__mnt_writers));
637 dput(mnt->mnt_root);
638 free_vfsmnt(mnt);
639 deactivate_super(sb);
642 void mntput_no_expire(struct vfsmount *mnt)
644 repeat:
645 if (atomic_dec_and_lock(&mnt->mnt_count, &vfsmount_lock)) {
646 if (likely(!mnt->mnt_pinned)) {
647 spin_unlock(&vfsmount_lock);
648 __mntput(mnt);
649 return;
651 atomic_add(mnt->mnt_pinned + 1, &mnt->mnt_count);
652 mnt->mnt_pinned = 0;
653 spin_unlock(&vfsmount_lock);
654 acct_auto_close_mnt(mnt);
655 security_sb_umount_close(mnt);
656 goto repeat;
660 EXPORT_SYMBOL(mntput_no_expire);
662 void mnt_pin(struct vfsmount *mnt)
664 spin_lock(&vfsmount_lock);
665 mnt->mnt_pinned++;
666 spin_unlock(&vfsmount_lock);
669 EXPORT_SYMBOL(mnt_pin);
671 void mnt_unpin(struct vfsmount *mnt)
673 spin_lock(&vfsmount_lock);
674 if (mnt->mnt_pinned) {
675 atomic_inc(&mnt->mnt_count);
676 mnt->mnt_pinned--;
678 spin_unlock(&vfsmount_lock);
681 EXPORT_SYMBOL(mnt_unpin);
683 static inline void mangle(struct seq_file *m, const char *s)
685 seq_escape(m, s, " \t\n\\");
689 * Simple .show_options callback for filesystems which don't want to
690 * implement more complex mount option showing.
692 * See also save_mount_options().
694 int generic_show_options(struct seq_file *m, struct vfsmount *mnt)
696 const char *options = mnt->mnt_sb->s_options;
698 if (options != NULL && options[0]) {
699 seq_putc(m, ',');
700 mangle(m, options);
703 return 0;
705 EXPORT_SYMBOL(generic_show_options);
708 * If filesystem uses generic_show_options(), this function should be
709 * called from the fill_super() callback.
711 * The .remount_fs callback usually needs to be handled in a special
712 * way, to make sure, that previous options are not overwritten if the
713 * remount fails.
715 * Also note, that if the filesystem's .remount_fs function doesn't
716 * reset all options to their default value, but changes only newly
717 * given options, then the displayed options will not reflect reality
718 * any more.
720 void save_mount_options(struct super_block *sb, char *options)
722 kfree(sb->s_options);
723 sb->s_options = kstrdup(options, GFP_KERNEL);
725 EXPORT_SYMBOL(save_mount_options);
727 #ifdef CONFIG_PROC_FS
728 /* iterator */
729 static void *m_start(struct seq_file *m, loff_t *pos)
731 struct proc_mounts *p = m->private;
733 down_read(&namespace_sem);
734 return seq_list_start(&p->ns->list, *pos);
737 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
739 struct proc_mounts *p = m->private;
741 return seq_list_next(v, &p->ns->list, pos);
744 static void m_stop(struct seq_file *m, void *v)
746 up_read(&namespace_sem);
749 struct proc_fs_info {
750 int flag;
751 const char *str;
754 static void show_sb_opts(struct seq_file *m, struct super_block *sb)
756 static const struct proc_fs_info fs_info[] = {
757 { MS_SYNCHRONOUS, ",sync" },
758 { MS_DIRSYNC, ",dirsync" },
759 { MS_MANDLOCK, ",mand" },
760 { 0, NULL }
762 const struct proc_fs_info *fs_infop;
764 for (fs_infop = fs_info; fs_infop->flag; fs_infop++) {
765 if (sb->s_flags & fs_infop->flag)
766 seq_puts(m, fs_infop->str);
770 static void show_mnt_opts(struct seq_file *m, struct vfsmount *mnt)
772 static const struct proc_fs_info mnt_info[] = {
773 { MNT_NOSUID, ",nosuid" },
774 { MNT_NODEV, ",nodev" },
775 { MNT_NOEXEC, ",noexec" },
776 { MNT_NOATIME, ",noatime" },
777 { MNT_NODIRATIME, ",nodiratime" },
778 { MNT_RELATIME, ",relatime" },
779 { 0, NULL }
781 const struct proc_fs_info *fs_infop;
783 for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) {
784 if (mnt->mnt_flags & fs_infop->flag)
785 seq_puts(m, fs_infop->str);
789 static void show_type(struct seq_file *m, struct super_block *sb)
791 mangle(m, sb->s_type->name);
792 if (sb->s_subtype && sb->s_subtype[0]) {
793 seq_putc(m, '.');
794 mangle(m, sb->s_subtype);
798 static int show_vfsmnt(struct seq_file *m, void *v)
800 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
801 int err = 0;
802 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
804 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
805 seq_putc(m, ' ');
806 seq_path(m, &mnt_path, " \t\n\\");
807 seq_putc(m, ' ');
808 show_type(m, mnt->mnt_sb);
809 seq_puts(m, __mnt_is_readonly(mnt) ? " ro" : " rw");
810 show_sb_opts(m, mnt->mnt_sb);
811 show_mnt_opts(m, mnt);
812 if (mnt->mnt_sb->s_op->show_options)
813 err = mnt->mnt_sb->s_op->show_options(m, mnt);
814 seq_puts(m, " 0 0\n");
815 return err;
818 const struct seq_operations mounts_op = {
819 .start = m_start,
820 .next = m_next,
821 .stop = m_stop,
822 .show = show_vfsmnt
825 static int show_mountinfo(struct seq_file *m, void *v)
827 struct proc_mounts *p = m->private;
828 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
829 struct super_block *sb = mnt->mnt_sb;
830 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
831 struct path root = p->root;
832 int err = 0;
834 seq_printf(m, "%i %i %u:%u ", mnt->mnt_id, mnt->mnt_parent->mnt_id,
835 MAJOR(sb->s_dev), MINOR(sb->s_dev));
836 seq_dentry(m, mnt->mnt_root, " \t\n\\");
837 seq_putc(m, ' ');
838 seq_path_root(m, &mnt_path, &root, " \t\n\\");
839 if (root.mnt != p->root.mnt || root.dentry != p->root.dentry) {
841 * Mountpoint is outside root, discard that one. Ugly,
842 * but less so than trying to do that in iterator in a
843 * race-free way (due to renames).
845 return SEQ_SKIP;
847 seq_puts(m, mnt->mnt_flags & MNT_READONLY ? " ro" : " rw");
848 show_mnt_opts(m, mnt);
850 /* Tagged fields ("foo:X" or "bar") */
851 if (IS_MNT_SHARED(mnt))
852 seq_printf(m, " shared:%i", mnt->mnt_group_id);
853 if (IS_MNT_SLAVE(mnt)) {
854 int master = mnt->mnt_master->mnt_group_id;
855 int dom = get_dominating_id(mnt, &p->root);
856 seq_printf(m, " master:%i", master);
857 if (dom && dom != master)
858 seq_printf(m, " propagate_from:%i", dom);
860 if (IS_MNT_UNBINDABLE(mnt))
861 seq_puts(m, " unbindable");
863 /* Filesystem specific data */
864 seq_puts(m, " - ");
865 show_type(m, sb);
866 seq_putc(m, ' ');
867 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
868 seq_puts(m, sb->s_flags & MS_RDONLY ? " ro" : " rw");
869 show_sb_opts(m, sb);
870 if (sb->s_op->show_options)
871 err = sb->s_op->show_options(m, mnt);
872 seq_putc(m, '\n');
873 return err;
876 const struct seq_operations mountinfo_op = {
877 .start = m_start,
878 .next = m_next,
879 .stop = m_stop,
880 .show = show_mountinfo,
883 static int show_vfsstat(struct seq_file *m, void *v)
885 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
886 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
887 int err = 0;
889 /* device */
890 if (mnt->mnt_devname) {
891 seq_puts(m, "device ");
892 mangle(m, mnt->mnt_devname);
893 } else
894 seq_puts(m, "no device");
896 /* mount point */
897 seq_puts(m, " mounted on ");
898 seq_path(m, &mnt_path, " \t\n\\");
899 seq_putc(m, ' ');
901 /* file system type */
902 seq_puts(m, "with fstype ");
903 show_type(m, mnt->mnt_sb);
905 /* optional statistics */
906 if (mnt->mnt_sb->s_op->show_stats) {
907 seq_putc(m, ' ');
908 err = mnt->mnt_sb->s_op->show_stats(m, mnt);
911 seq_putc(m, '\n');
912 return err;
915 const struct seq_operations mountstats_op = {
916 .start = m_start,
917 .next = m_next,
918 .stop = m_stop,
919 .show = show_vfsstat,
921 #endif /* CONFIG_PROC_FS */
924 * may_umount_tree - check if a mount tree is busy
925 * @mnt: root of mount tree
927 * This is called to check if a tree of mounts has any
928 * open files, pwds, chroots or sub mounts that are
929 * busy.
931 int may_umount_tree(struct vfsmount *mnt)
933 int actual_refs = 0;
934 int minimum_refs = 0;
935 struct vfsmount *p;
937 spin_lock(&vfsmount_lock);
938 for (p = mnt; p; p = next_mnt(p, mnt)) {
939 actual_refs += atomic_read(&p->mnt_count);
940 minimum_refs += 2;
942 spin_unlock(&vfsmount_lock);
944 if (actual_refs > minimum_refs)
945 return 0;
947 return 1;
950 EXPORT_SYMBOL(may_umount_tree);
953 * may_umount - check if a mount point is busy
954 * @mnt: root of mount
956 * This is called to check if a mount point has any
957 * open files, pwds, chroots or sub mounts. If the
958 * mount has sub mounts this will return busy
959 * regardless of whether the sub mounts are busy.
961 * Doesn't take quota and stuff into account. IOW, in some cases it will
962 * give false negatives. The main reason why it's here is that we need
963 * a non-destructive way to look for easily umountable filesystems.
965 int may_umount(struct vfsmount *mnt)
967 int ret = 1;
968 spin_lock(&vfsmount_lock);
969 if (propagate_mount_busy(mnt, 2))
970 ret = 0;
971 spin_unlock(&vfsmount_lock);
972 return ret;
975 EXPORT_SYMBOL(may_umount);
977 void release_mounts(struct list_head *head)
979 struct vfsmount *mnt;
980 while (!list_empty(head)) {
981 mnt = list_first_entry(head, struct vfsmount, mnt_hash);
982 list_del_init(&mnt->mnt_hash);
983 if (mnt->mnt_parent != mnt) {
984 struct dentry *dentry;
985 struct vfsmount *m;
986 spin_lock(&vfsmount_lock);
987 dentry = mnt->mnt_mountpoint;
988 m = mnt->mnt_parent;
989 mnt->mnt_mountpoint = mnt->mnt_root;
990 mnt->mnt_parent = mnt;
991 m->mnt_ghosts--;
992 spin_unlock(&vfsmount_lock);
993 dput(dentry);
994 mntput(m);
996 mntput(mnt);
1000 void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill)
1002 struct vfsmount *p;
1004 for (p = mnt; p; p = next_mnt(p, mnt))
1005 list_move(&p->mnt_hash, kill);
1007 if (propagate)
1008 propagate_umount(kill);
1010 list_for_each_entry(p, kill, mnt_hash) {
1011 list_del_init(&p->mnt_expire);
1012 list_del_init(&p->mnt_list);
1013 __touch_mnt_namespace(p->mnt_ns);
1014 p->mnt_ns = NULL;
1015 list_del_init(&p->mnt_child);
1016 if (p->mnt_parent != p) {
1017 p->mnt_parent->mnt_ghosts++;
1018 p->mnt_mountpoint->d_mounted--;
1020 change_mnt_propagation(p, MS_PRIVATE);
1024 static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts);
1026 static int do_umount(struct vfsmount *mnt, int flags)
1028 struct super_block *sb = mnt->mnt_sb;
1029 int retval;
1030 LIST_HEAD(umount_list);
1032 retval = security_sb_umount(mnt, flags);
1033 if (retval)
1034 return retval;
1037 * Allow userspace to request a mountpoint be expired rather than
1038 * unmounting unconditionally. Unmount only happens if:
1039 * (1) the mark is already set (the mark is cleared by mntput())
1040 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1042 if (flags & MNT_EXPIRE) {
1043 if (mnt == current->fs->root.mnt ||
1044 flags & (MNT_FORCE | MNT_DETACH))
1045 return -EINVAL;
1047 if (atomic_read(&mnt->mnt_count) != 2)
1048 return -EBUSY;
1050 if (!xchg(&mnt->mnt_expiry_mark, 1))
1051 return -EAGAIN;
1055 * If we may have to abort operations to get out of this
1056 * mount, and they will themselves hold resources we must
1057 * allow the fs to do things. In the Unix tradition of
1058 * 'Gee thats tricky lets do it in userspace' the umount_begin
1059 * might fail to complete on the first run through as other tasks
1060 * must return, and the like. Thats for the mount program to worry
1061 * about for the moment.
1064 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1065 lock_kernel();
1066 sb->s_op->umount_begin(sb);
1067 unlock_kernel();
1071 * No sense to grab the lock for this test, but test itself looks
1072 * somewhat bogus. Suggestions for better replacement?
1073 * Ho-hum... In principle, we might treat that as umount + switch
1074 * to rootfs. GC would eventually take care of the old vfsmount.
1075 * Actually it makes sense, especially if rootfs would contain a
1076 * /reboot - static binary that would close all descriptors and
1077 * call reboot(9). Then init(8) could umount root and exec /reboot.
1079 if (mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1081 * Special case for "unmounting" root ...
1082 * we just try to remount it readonly.
1084 down_write(&sb->s_umount);
1085 if (!(sb->s_flags & MS_RDONLY)) {
1086 lock_kernel();
1087 DQUOT_OFF(sb);
1088 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
1089 unlock_kernel();
1091 up_write(&sb->s_umount);
1092 return retval;
1095 down_write(&namespace_sem);
1096 spin_lock(&vfsmount_lock);
1097 event++;
1099 if (!(flags & MNT_DETACH))
1100 shrink_submounts(mnt, &umount_list);
1102 retval = -EBUSY;
1103 if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
1104 if (!list_empty(&mnt->mnt_list))
1105 umount_tree(mnt, 1, &umount_list);
1106 retval = 0;
1108 spin_unlock(&vfsmount_lock);
1109 if (retval)
1110 security_sb_umount_busy(mnt);
1111 up_write(&namespace_sem);
1112 release_mounts(&umount_list);
1113 return retval;
1117 * Now umount can handle mount points as well as block devices.
1118 * This is important for filesystems which use unnamed block devices.
1120 * We now support a flag for forced unmount like the other 'big iron'
1121 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1124 asmlinkage long sys_umount(char __user * name, int flags)
1126 struct nameidata nd;
1127 int retval;
1129 retval = __user_walk(name, LOOKUP_FOLLOW, &nd);
1130 if (retval)
1131 goto out;
1132 retval = -EINVAL;
1133 if (nd.path.dentry != nd.path.mnt->mnt_root)
1134 goto dput_and_out;
1135 if (!check_mnt(nd.path.mnt))
1136 goto dput_and_out;
1138 retval = -EPERM;
1139 if (!capable(CAP_SYS_ADMIN))
1140 goto dput_and_out;
1142 retval = do_umount(nd.path.mnt, flags);
1143 dput_and_out:
1144 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1145 dput(nd.path.dentry);
1146 mntput_no_expire(nd.path.mnt);
1147 out:
1148 return retval;
1151 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1154 * The 2.0 compatible umount. No flags.
1156 asmlinkage long sys_oldumount(char __user * name)
1158 return sys_umount(name, 0);
1161 #endif
1163 static int mount_is_safe(struct nameidata *nd)
1165 if (capable(CAP_SYS_ADMIN))
1166 return 0;
1167 return -EPERM;
1168 #ifdef notyet
1169 if (S_ISLNK(nd->path.dentry->d_inode->i_mode))
1170 return -EPERM;
1171 if (nd->path.dentry->d_inode->i_mode & S_ISVTX) {
1172 if (current->uid != nd->path.dentry->d_inode->i_uid)
1173 return -EPERM;
1175 if (vfs_permission(nd, MAY_WRITE))
1176 return -EPERM;
1177 return 0;
1178 #endif
1181 static int lives_below_in_same_fs(struct dentry *d, struct dentry *dentry)
1183 while (1) {
1184 if (d == dentry)
1185 return 1;
1186 if (d == NULL || d == d->d_parent)
1187 return 0;
1188 d = d->d_parent;
1192 struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry,
1193 int flag)
1195 struct vfsmount *res, *p, *q, *r, *s;
1196 struct path path;
1198 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
1199 return NULL;
1201 res = q = clone_mnt(mnt, dentry, flag);
1202 if (!q)
1203 goto Enomem;
1204 q->mnt_mountpoint = mnt->mnt_mountpoint;
1206 p = mnt;
1207 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1208 if (!lives_below_in_same_fs(r->mnt_mountpoint, dentry))
1209 continue;
1211 for (s = r; s; s = next_mnt(s, r)) {
1212 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) {
1213 s = skip_mnt_tree(s);
1214 continue;
1216 while (p != s->mnt_parent) {
1217 p = p->mnt_parent;
1218 q = q->mnt_parent;
1220 p = s;
1221 path.mnt = q;
1222 path.dentry = p->mnt_mountpoint;
1223 q = clone_mnt(p, p->mnt_root, flag);
1224 if (!q)
1225 goto Enomem;
1226 spin_lock(&vfsmount_lock);
1227 list_add_tail(&q->mnt_list, &res->mnt_list);
1228 attach_mnt(q, &path);
1229 spin_unlock(&vfsmount_lock);
1232 return res;
1233 Enomem:
1234 if (res) {
1235 LIST_HEAD(umount_list);
1236 spin_lock(&vfsmount_lock);
1237 umount_tree(res, 0, &umount_list);
1238 spin_unlock(&vfsmount_lock);
1239 release_mounts(&umount_list);
1241 return NULL;
1244 struct vfsmount *collect_mounts(struct vfsmount *mnt, struct dentry *dentry)
1246 struct vfsmount *tree;
1247 down_write(&namespace_sem);
1248 tree = copy_tree(mnt, dentry, CL_COPY_ALL | CL_PRIVATE);
1249 up_write(&namespace_sem);
1250 return tree;
1253 void drop_collected_mounts(struct vfsmount *mnt)
1255 LIST_HEAD(umount_list);
1256 down_write(&namespace_sem);
1257 spin_lock(&vfsmount_lock);
1258 umount_tree(mnt, 0, &umount_list);
1259 spin_unlock(&vfsmount_lock);
1260 up_write(&namespace_sem);
1261 release_mounts(&umount_list);
1264 static void cleanup_group_ids(struct vfsmount *mnt, struct vfsmount *end)
1266 struct vfsmount *p;
1268 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1269 if (p->mnt_group_id && !IS_MNT_SHARED(p))
1270 mnt_release_group_id(p);
1274 static int invent_group_ids(struct vfsmount *mnt, bool recurse)
1276 struct vfsmount *p;
1278 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1279 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1280 int err = mnt_alloc_group_id(p);
1281 if (err) {
1282 cleanup_group_ids(mnt, p);
1283 return err;
1288 return 0;
1292 * @source_mnt : mount tree to be attached
1293 * @nd : place the mount tree @source_mnt is attached
1294 * @parent_nd : if non-null, detach the source_mnt from its parent and
1295 * store the parent mount and mountpoint dentry.
1296 * (done when source_mnt is moved)
1298 * NOTE: in the table below explains the semantics when a source mount
1299 * of a given type is attached to a destination mount of a given type.
1300 * ---------------------------------------------------------------------------
1301 * | BIND MOUNT OPERATION |
1302 * |**************************************************************************
1303 * | source-->| shared | private | slave | unbindable |
1304 * | dest | | | | |
1305 * | | | | | | |
1306 * | v | | | | |
1307 * |**************************************************************************
1308 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1309 * | | | | | |
1310 * |non-shared| shared (+) | private | slave (*) | invalid |
1311 * ***************************************************************************
1312 * A bind operation clones the source mount and mounts the clone on the
1313 * destination mount.
1315 * (++) the cloned mount is propagated to all the mounts in the propagation
1316 * tree of the destination mount and the cloned mount is added to
1317 * the peer group of the source mount.
1318 * (+) the cloned mount is created under the destination mount and is marked
1319 * as shared. The cloned mount is added to the peer group of the source
1320 * mount.
1321 * (+++) the mount is propagated to all the mounts in the propagation tree
1322 * of the destination mount and the cloned mount is made slave
1323 * of the same master as that of the source mount. The cloned mount
1324 * is marked as 'shared and slave'.
1325 * (*) the cloned mount is made a slave of the same master as that of the
1326 * source mount.
1328 * ---------------------------------------------------------------------------
1329 * | MOVE MOUNT OPERATION |
1330 * |**************************************************************************
1331 * | source-->| shared | private | slave | unbindable |
1332 * | dest | | | | |
1333 * | | | | | | |
1334 * | v | | | | |
1335 * |**************************************************************************
1336 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1337 * | | | | | |
1338 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1339 * ***************************************************************************
1341 * (+) the mount is moved to the destination. And is then propagated to
1342 * all the mounts in the propagation tree of the destination mount.
1343 * (+*) the mount is moved to the destination.
1344 * (+++) the mount is moved to the destination and is then propagated to
1345 * all the mounts belonging to the destination mount's propagation tree.
1346 * the mount is marked as 'shared and slave'.
1347 * (*) the mount continues to be a slave at the new location.
1349 * if the source mount is a tree, the operations explained above is
1350 * applied to each mount in the tree.
1351 * Must be called without spinlocks held, since this function can sleep
1352 * in allocations.
1354 static int attach_recursive_mnt(struct vfsmount *source_mnt,
1355 struct path *path, struct path *parent_path)
1357 LIST_HEAD(tree_list);
1358 struct vfsmount *dest_mnt = path->mnt;
1359 struct dentry *dest_dentry = path->dentry;
1360 struct vfsmount *child, *p;
1361 int err;
1363 if (IS_MNT_SHARED(dest_mnt)) {
1364 err = invent_group_ids(source_mnt, true);
1365 if (err)
1366 goto out;
1368 err = propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list);
1369 if (err)
1370 goto out_cleanup_ids;
1372 if (IS_MNT_SHARED(dest_mnt)) {
1373 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
1374 set_mnt_shared(p);
1377 spin_lock(&vfsmount_lock);
1378 if (parent_path) {
1379 detach_mnt(source_mnt, parent_path);
1380 attach_mnt(source_mnt, path);
1381 touch_mnt_namespace(current->nsproxy->mnt_ns);
1382 } else {
1383 mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
1384 commit_tree(source_mnt);
1387 list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
1388 list_del_init(&child->mnt_hash);
1389 commit_tree(child);
1391 spin_unlock(&vfsmount_lock);
1392 return 0;
1394 out_cleanup_ids:
1395 if (IS_MNT_SHARED(dest_mnt))
1396 cleanup_group_ids(source_mnt, NULL);
1397 out:
1398 return err;
1401 static int graft_tree(struct vfsmount *mnt, struct path *path)
1403 int err;
1404 if (mnt->mnt_sb->s_flags & MS_NOUSER)
1405 return -EINVAL;
1407 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1408 S_ISDIR(mnt->mnt_root->d_inode->i_mode))
1409 return -ENOTDIR;
1411 err = -ENOENT;
1412 mutex_lock(&path->dentry->d_inode->i_mutex);
1413 if (IS_DEADDIR(path->dentry->d_inode))
1414 goto out_unlock;
1416 err = security_sb_check_sb(mnt, path);
1417 if (err)
1418 goto out_unlock;
1420 err = -ENOENT;
1421 if (IS_ROOT(path->dentry) || !d_unhashed(path->dentry))
1422 err = attach_recursive_mnt(mnt, path, NULL);
1423 out_unlock:
1424 mutex_unlock(&path->dentry->d_inode->i_mutex);
1425 if (!err)
1426 security_sb_post_addmount(mnt, path);
1427 return err;
1431 * recursively change the type of the mountpoint.
1432 * noinline this do_mount helper to save do_mount stack space.
1434 static noinline int do_change_type(struct nameidata *nd, int flag)
1436 struct vfsmount *m, *mnt = nd->path.mnt;
1437 int recurse = flag & MS_REC;
1438 int type = flag & ~MS_REC;
1439 int err = 0;
1441 if (!capable(CAP_SYS_ADMIN))
1442 return -EPERM;
1444 if (nd->path.dentry != nd->path.mnt->mnt_root)
1445 return -EINVAL;
1447 down_write(&namespace_sem);
1448 if (type == MS_SHARED) {
1449 err = invent_group_ids(mnt, recurse);
1450 if (err)
1451 goto out_unlock;
1454 spin_lock(&vfsmount_lock);
1455 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
1456 change_mnt_propagation(m, type);
1457 spin_unlock(&vfsmount_lock);
1459 out_unlock:
1460 up_write(&namespace_sem);
1461 return err;
1465 * do loopback mount.
1466 * noinline this do_mount helper to save do_mount stack space.
1468 static noinline int do_loopback(struct nameidata *nd, char *old_name,
1469 int recurse)
1471 struct nameidata old_nd;
1472 struct vfsmount *mnt = NULL;
1473 int err = mount_is_safe(nd);
1474 if (err)
1475 return err;
1476 if (!old_name || !*old_name)
1477 return -EINVAL;
1478 err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd);
1479 if (err)
1480 return err;
1482 down_write(&namespace_sem);
1483 err = -EINVAL;
1484 if (IS_MNT_UNBINDABLE(old_nd.path.mnt))
1485 goto out;
1487 if (!check_mnt(nd->path.mnt) || !check_mnt(old_nd.path.mnt))
1488 goto out;
1490 err = -ENOMEM;
1491 if (recurse)
1492 mnt = copy_tree(old_nd.path.mnt, old_nd.path.dentry, 0);
1493 else
1494 mnt = clone_mnt(old_nd.path.mnt, old_nd.path.dentry, 0);
1496 if (!mnt)
1497 goto out;
1499 err = graft_tree(mnt, &nd->path);
1500 if (err) {
1501 LIST_HEAD(umount_list);
1502 spin_lock(&vfsmount_lock);
1503 umount_tree(mnt, 0, &umount_list);
1504 spin_unlock(&vfsmount_lock);
1505 release_mounts(&umount_list);
1508 out:
1509 up_write(&namespace_sem);
1510 path_put(&old_nd.path);
1511 return err;
1514 static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
1516 int error = 0;
1517 int readonly_request = 0;
1519 if (ms_flags & MS_RDONLY)
1520 readonly_request = 1;
1521 if (readonly_request == __mnt_is_readonly(mnt))
1522 return 0;
1524 if (readonly_request)
1525 error = mnt_make_readonly(mnt);
1526 else
1527 __mnt_unmake_readonly(mnt);
1528 return error;
1532 * change filesystem flags. dir should be a physical root of filesystem.
1533 * If you've mounted a non-root directory somewhere and want to do remount
1534 * on it - tough luck.
1535 * noinline this do_mount helper to save do_mount stack space.
1537 static noinline int do_remount(struct nameidata *nd, int flags, int mnt_flags,
1538 void *data)
1540 int err;
1541 struct super_block *sb = nd->path.mnt->mnt_sb;
1543 if (!capable(CAP_SYS_ADMIN))
1544 return -EPERM;
1546 if (!check_mnt(nd->path.mnt))
1547 return -EINVAL;
1549 if (nd->path.dentry != nd->path.mnt->mnt_root)
1550 return -EINVAL;
1552 down_write(&sb->s_umount);
1553 if (flags & MS_BIND)
1554 err = change_mount_flags(nd->path.mnt, flags);
1555 else
1556 err = do_remount_sb(sb, flags, data, 0);
1557 if (!err)
1558 nd->path.mnt->mnt_flags = mnt_flags;
1559 up_write(&sb->s_umount);
1560 if (!err)
1561 security_sb_post_remount(nd->path.mnt, flags, data);
1562 return err;
1565 static inline int tree_contains_unbindable(struct vfsmount *mnt)
1567 struct vfsmount *p;
1568 for (p = mnt; p; p = next_mnt(p, mnt)) {
1569 if (IS_MNT_UNBINDABLE(p))
1570 return 1;
1572 return 0;
1576 * noinline this do_mount helper to save do_mount stack space.
1578 static noinline int do_move_mount(struct nameidata *nd, char *old_name)
1580 struct nameidata old_nd;
1581 struct path parent_path;
1582 struct vfsmount *p;
1583 int err = 0;
1584 if (!capable(CAP_SYS_ADMIN))
1585 return -EPERM;
1586 if (!old_name || !*old_name)
1587 return -EINVAL;
1588 err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd);
1589 if (err)
1590 return err;
1592 down_write(&namespace_sem);
1593 while (d_mountpoint(nd->path.dentry) &&
1594 follow_down(&nd->path.mnt, &nd->path.dentry))
1596 err = -EINVAL;
1597 if (!check_mnt(nd->path.mnt) || !check_mnt(old_nd.path.mnt))
1598 goto out;
1600 err = -ENOENT;
1601 mutex_lock(&nd->path.dentry->d_inode->i_mutex);
1602 if (IS_DEADDIR(nd->path.dentry->d_inode))
1603 goto out1;
1605 if (!IS_ROOT(nd->path.dentry) && d_unhashed(nd->path.dentry))
1606 goto out1;
1608 err = -EINVAL;
1609 if (old_nd.path.dentry != old_nd.path.mnt->mnt_root)
1610 goto out1;
1612 if (old_nd.path.mnt == old_nd.path.mnt->mnt_parent)
1613 goto out1;
1615 if (S_ISDIR(nd->path.dentry->d_inode->i_mode) !=
1616 S_ISDIR(old_nd.path.dentry->d_inode->i_mode))
1617 goto out1;
1619 * Don't move a mount residing in a shared parent.
1621 if (old_nd.path.mnt->mnt_parent &&
1622 IS_MNT_SHARED(old_nd.path.mnt->mnt_parent))
1623 goto out1;
1625 * Don't move a mount tree containing unbindable mounts to a destination
1626 * mount which is shared.
1628 if (IS_MNT_SHARED(nd->path.mnt) &&
1629 tree_contains_unbindable(old_nd.path.mnt))
1630 goto out1;
1631 err = -ELOOP;
1632 for (p = nd->path.mnt; p->mnt_parent != p; p = p->mnt_parent)
1633 if (p == old_nd.path.mnt)
1634 goto out1;
1636 err = attach_recursive_mnt(old_nd.path.mnt, &nd->path, &parent_path);
1637 if (err)
1638 goto out1;
1640 /* if the mount is moved, it should no longer be expire
1641 * automatically */
1642 list_del_init(&old_nd.path.mnt->mnt_expire);
1643 out1:
1644 mutex_unlock(&nd->path.dentry->d_inode->i_mutex);
1645 out:
1646 up_write(&namespace_sem);
1647 if (!err)
1648 path_put(&parent_path);
1649 path_put(&old_nd.path);
1650 return err;
1654 * create a new mount for userspace and request it to be added into the
1655 * namespace's tree
1656 * noinline this do_mount helper to save do_mount stack space.
1658 static noinline int do_new_mount(struct nameidata *nd, char *type, int flags,
1659 int mnt_flags, char *name, void *data)
1661 struct vfsmount *mnt;
1663 if (!type || !memchr(type, 0, PAGE_SIZE))
1664 return -EINVAL;
1666 /* we need capabilities... */
1667 if (!capable(CAP_SYS_ADMIN))
1668 return -EPERM;
1670 mnt = do_kern_mount(type, flags, name, data);
1671 if (IS_ERR(mnt))
1672 return PTR_ERR(mnt);
1674 return do_add_mount(mnt, nd, mnt_flags, NULL);
1678 * add a mount into a namespace's mount tree
1679 * - provide the option of adding the new mount to an expiration list
1681 int do_add_mount(struct vfsmount *newmnt, struct nameidata *nd,
1682 int mnt_flags, struct list_head *fslist)
1684 int err;
1686 down_write(&namespace_sem);
1687 /* Something was mounted here while we slept */
1688 while (d_mountpoint(nd->path.dentry) &&
1689 follow_down(&nd->path.mnt, &nd->path.dentry))
1691 err = -EINVAL;
1692 if (!check_mnt(nd->path.mnt))
1693 goto unlock;
1695 /* Refuse the same filesystem on the same mount point */
1696 err = -EBUSY;
1697 if (nd->path.mnt->mnt_sb == newmnt->mnt_sb &&
1698 nd->path.mnt->mnt_root == nd->path.dentry)
1699 goto unlock;
1701 err = -EINVAL;
1702 if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode))
1703 goto unlock;
1705 newmnt->mnt_flags = mnt_flags;
1706 if ((err = graft_tree(newmnt, &nd->path)))
1707 goto unlock;
1709 if (fslist) /* add to the specified expiration list */
1710 list_add_tail(&newmnt->mnt_expire, fslist);
1712 up_write(&namespace_sem);
1713 return 0;
1715 unlock:
1716 up_write(&namespace_sem);
1717 mntput(newmnt);
1718 return err;
1721 EXPORT_SYMBOL_GPL(do_add_mount);
1724 * process a list of expirable mountpoints with the intent of discarding any
1725 * mountpoints that aren't in use and haven't been touched since last we came
1726 * here
1728 void mark_mounts_for_expiry(struct list_head *mounts)
1730 struct vfsmount *mnt, *next;
1731 LIST_HEAD(graveyard);
1732 LIST_HEAD(umounts);
1734 if (list_empty(mounts))
1735 return;
1737 down_write(&namespace_sem);
1738 spin_lock(&vfsmount_lock);
1740 /* extract from the expiration list every vfsmount that matches the
1741 * following criteria:
1742 * - only referenced by its parent vfsmount
1743 * - still marked for expiry (marked on the last call here; marks are
1744 * cleared by mntput())
1746 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
1747 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
1748 propagate_mount_busy(mnt, 1))
1749 continue;
1750 list_move(&mnt->mnt_expire, &graveyard);
1752 while (!list_empty(&graveyard)) {
1753 mnt = list_first_entry(&graveyard, struct vfsmount, mnt_expire);
1754 touch_mnt_namespace(mnt->mnt_ns);
1755 umount_tree(mnt, 1, &umounts);
1757 spin_unlock(&vfsmount_lock);
1758 up_write(&namespace_sem);
1760 release_mounts(&umounts);
1763 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
1766 * Ripoff of 'select_parent()'
1768 * search the list of submounts for a given mountpoint, and move any
1769 * shrinkable submounts to the 'graveyard' list.
1771 static int select_submounts(struct vfsmount *parent, struct list_head *graveyard)
1773 struct vfsmount *this_parent = parent;
1774 struct list_head *next;
1775 int found = 0;
1777 repeat:
1778 next = this_parent->mnt_mounts.next;
1779 resume:
1780 while (next != &this_parent->mnt_mounts) {
1781 struct list_head *tmp = next;
1782 struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child);
1784 next = tmp->next;
1785 if (!(mnt->mnt_flags & MNT_SHRINKABLE))
1786 continue;
1788 * Descend a level if the d_mounts list is non-empty.
1790 if (!list_empty(&mnt->mnt_mounts)) {
1791 this_parent = mnt;
1792 goto repeat;
1795 if (!propagate_mount_busy(mnt, 1)) {
1796 list_move_tail(&mnt->mnt_expire, graveyard);
1797 found++;
1801 * All done at this level ... ascend and resume the search
1803 if (this_parent != parent) {
1804 next = this_parent->mnt_child.next;
1805 this_parent = this_parent->mnt_parent;
1806 goto resume;
1808 return found;
1812 * process a list of expirable mountpoints with the intent of discarding any
1813 * submounts of a specific parent mountpoint
1815 static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts)
1817 LIST_HEAD(graveyard);
1818 struct vfsmount *m;
1820 /* extract submounts of 'mountpoint' from the expiration list */
1821 while (select_submounts(mnt, &graveyard)) {
1822 while (!list_empty(&graveyard)) {
1823 m = list_first_entry(&graveyard, struct vfsmount,
1824 mnt_expire);
1825 touch_mnt_namespace(mnt->mnt_ns);
1826 umount_tree(mnt, 1, umounts);
1832 * Some copy_from_user() implementations do not return the exact number of
1833 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
1834 * Note that this function differs from copy_from_user() in that it will oops
1835 * on bad values of `to', rather than returning a short copy.
1837 static long exact_copy_from_user(void *to, const void __user * from,
1838 unsigned long n)
1840 char *t = to;
1841 const char __user *f = from;
1842 char c;
1844 if (!access_ok(VERIFY_READ, from, n))
1845 return n;
1847 while (n) {
1848 if (__get_user(c, f)) {
1849 memset(t, 0, n);
1850 break;
1852 *t++ = c;
1853 f++;
1854 n--;
1856 return n;
1859 int copy_mount_options(const void __user * data, unsigned long *where)
1861 int i;
1862 unsigned long page;
1863 unsigned long size;
1865 *where = 0;
1866 if (!data)
1867 return 0;
1869 if (!(page = __get_free_page(GFP_KERNEL)))
1870 return -ENOMEM;
1872 /* We only care that *some* data at the address the user
1873 * gave us is valid. Just in case, we'll zero
1874 * the remainder of the page.
1876 /* copy_from_user cannot cross TASK_SIZE ! */
1877 size = TASK_SIZE - (unsigned long)data;
1878 if (size > PAGE_SIZE)
1879 size = PAGE_SIZE;
1881 i = size - exact_copy_from_user((void *)page, data, size);
1882 if (!i) {
1883 free_page(page);
1884 return -EFAULT;
1886 if (i != PAGE_SIZE)
1887 memset((char *)page + i, 0, PAGE_SIZE - i);
1888 *where = page;
1889 return 0;
1893 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
1894 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
1896 * data is a (void *) that can point to any structure up to
1897 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
1898 * information (or be NULL).
1900 * Pre-0.97 versions of mount() didn't have a flags word.
1901 * When the flags word was introduced its top half was required
1902 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
1903 * Therefore, if this magic number is present, it carries no information
1904 * and must be discarded.
1906 long do_mount(char *dev_name, char *dir_name, char *type_page,
1907 unsigned long flags, void *data_page)
1909 struct nameidata nd;
1910 int retval = 0;
1911 int mnt_flags = 0;
1913 /* Discard magic */
1914 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
1915 flags &= ~MS_MGC_MSK;
1917 /* Basic sanity checks */
1919 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
1920 return -EINVAL;
1921 if (dev_name && !memchr(dev_name, 0, PAGE_SIZE))
1922 return -EINVAL;
1924 if (data_page)
1925 ((char *)data_page)[PAGE_SIZE - 1] = 0;
1927 /* Separate the per-mountpoint flags */
1928 if (flags & MS_NOSUID)
1929 mnt_flags |= MNT_NOSUID;
1930 if (flags & MS_NODEV)
1931 mnt_flags |= MNT_NODEV;
1932 if (flags & MS_NOEXEC)
1933 mnt_flags |= MNT_NOEXEC;
1934 if (flags & MS_NOATIME)
1935 mnt_flags |= MNT_NOATIME;
1936 if (flags & MS_NODIRATIME)
1937 mnt_flags |= MNT_NODIRATIME;
1938 if (flags & MS_RELATIME)
1939 mnt_flags |= MNT_RELATIME;
1940 if (flags & MS_RDONLY)
1941 mnt_flags |= MNT_READONLY;
1943 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE |
1944 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT);
1946 /* ... and get the mountpoint */
1947 retval = path_lookup(dir_name, LOOKUP_FOLLOW, &nd);
1948 if (retval)
1949 return retval;
1951 retval = security_sb_mount(dev_name, &nd.path,
1952 type_page, flags, data_page);
1953 if (retval)
1954 goto dput_out;
1956 if (flags & MS_REMOUNT)
1957 retval = do_remount(&nd, flags & ~MS_REMOUNT, mnt_flags,
1958 data_page);
1959 else if (flags & MS_BIND)
1960 retval = do_loopback(&nd, dev_name, flags & MS_REC);
1961 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
1962 retval = do_change_type(&nd, flags);
1963 else if (flags & MS_MOVE)
1964 retval = do_move_mount(&nd, dev_name);
1965 else
1966 retval = do_new_mount(&nd, type_page, flags, mnt_flags,
1967 dev_name, data_page);
1968 dput_out:
1969 path_put(&nd.path);
1970 return retval;
1974 * Allocate a new namespace structure and populate it with contents
1975 * copied from the namespace of the passed in task structure.
1977 static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns,
1978 struct fs_struct *fs)
1980 struct mnt_namespace *new_ns;
1981 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL, *altrootmnt = NULL;
1982 struct vfsmount *p, *q;
1984 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
1985 if (!new_ns)
1986 return ERR_PTR(-ENOMEM);
1988 atomic_set(&new_ns->count, 1);
1989 INIT_LIST_HEAD(&new_ns->list);
1990 init_waitqueue_head(&new_ns->poll);
1991 new_ns->event = 0;
1993 down_write(&namespace_sem);
1994 /* First pass: copy the tree topology */
1995 new_ns->root = copy_tree(mnt_ns->root, mnt_ns->root->mnt_root,
1996 CL_COPY_ALL | CL_EXPIRE);
1997 if (!new_ns->root) {
1998 up_write(&namespace_sem);
1999 kfree(new_ns);
2000 return ERR_PTR(-ENOMEM);;
2002 spin_lock(&vfsmount_lock);
2003 list_add_tail(&new_ns->list, &new_ns->root->mnt_list);
2004 spin_unlock(&vfsmount_lock);
2007 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2008 * as belonging to new namespace. We have already acquired a private
2009 * fs_struct, so tsk->fs->lock is not needed.
2011 p = mnt_ns->root;
2012 q = new_ns->root;
2013 while (p) {
2014 q->mnt_ns = new_ns;
2015 if (fs) {
2016 if (p == fs->root.mnt) {
2017 rootmnt = p;
2018 fs->root.mnt = mntget(q);
2020 if (p == fs->pwd.mnt) {
2021 pwdmnt = p;
2022 fs->pwd.mnt = mntget(q);
2024 if (p == fs->altroot.mnt) {
2025 altrootmnt = p;
2026 fs->altroot.mnt = mntget(q);
2029 p = next_mnt(p, mnt_ns->root);
2030 q = next_mnt(q, new_ns->root);
2032 up_write(&namespace_sem);
2034 if (rootmnt)
2035 mntput(rootmnt);
2036 if (pwdmnt)
2037 mntput(pwdmnt);
2038 if (altrootmnt)
2039 mntput(altrootmnt);
2041 return new_ns;
2044 struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
2045 struct fs_struct *new_fs)
2047 struct mnt_namespace *new_ns;
2049 BUG_ON(!ns);
2050 get_mnt_ns(ns);
2052 if (!(flags & CLONE_NEWNS))
2053 return ns;
2055 new_ns = dup_mnt_ns(ns, new_fs);
2057 put_mnt_ns(ns);
2058 return new_ns;
2061 asmlinkage long sys_mount(char __user * dev_name, char __user * dir_name,
2062 char __user * type, unsigned long flags,
2063 void __user * data)
2065 int retval;
2066 unsigned long data_page;
2067 unsigned long type_page;
2068 unsigned long dev_page;
2069 char *dir_page;
2071 retval = copy_mount_options(type, &type_page);
2072 if (retval < 0)
2073 return retval;
2075 dir_page = getname(dir_name);
2076 retval = PTR_ERR(dir_page);
2077 if (IS_ERR(dir_page))
2078 goto out1;
2080 retval = copy_mount_options(dev_name, &dev_page);
2081 if (retval < 0)
2082 goto out2;
2084 retval = copy_mount_options(data, &data_page);
2085 if (retval < 0)
2086 goto out3;
2088 lock_kernel();
2089 retval = do_mount((char *)dev_page, dir_page, (char *)type_page,
2090 flags, (void *)data_page);
2091 unlock_kernel();
2092 free_page(data_page);
2094 out3:
2095 free_page(dev_page);
2096 out2:
2097 putname(dir_page);
2098 out1:
2099 free_page(type_page);
2100 return retval;
2104 * Replace the fs->{rootmnt,root} with {mnt,dentry}. Put the old values.
2105 * It can block. Requires the big lock held.
2107 void set_fs_root(struct fs_struct *fs, struct path *path)
2109 struct path old_root;
2111 write_lock(&fs->lock);
2112 old_root = fs->root;
2113 fs->root = *path;
2114 path_get(path);
2115 write_unlock(&fs->lock);
2116 if (old_root.dentry)
2117 path_put(&old_root);
2121 * Replace the fs->{pwdmnt,pwd} with {mnt,dentry}. Put the old values.
2122 * It can block. Requires the big lock held.
2124 void set_fs_pwd(struct fs_struct *fs, struct path *path)
2126 struct path old_pwd;
2128 write_lock(&fs->lock);
2129 old_pwd = fs->pwd;
2130 fs->pwd = *path;
2131 path_get(path);
2132 write_unlock(&fs->lock);
2134 if (old_pwd.dentry)
2135 path_put(&old_pwd);
2138 static void chroot_fs_refs(struct path *old_root, struct path *new_root)
2140 struct task_struct *g, *p;
2141 struct fs_struct *fs;
2143 read_lock(&tasklist_lock);
2144 do_each_thread(g, p) {
2145 task_lock(p);
2146 fs = p->fs;
2147 if (fs) {
2148 atomic_inc(&fs->count);
2149 task_unlock(p);
2150 if (fs->root.dentry == old_root->dentry
2151 && fs->root.mnt == old_root->mnt)
2152 set_fs_root(fs, new_root);
2153 if (fs->pwd.dentry == old_root->dentry
2154 && fs->pwd.mnt == old_root->mnt)
2155 set_fs_pwd(fs, new_root);
2156 put_fs_struct(fs);
2157 } else
2158 task_unlock(p);
2159 } while_each_thread(g, p);
2160 read_unlock(&tasklist_lock);
2164 * pivot_root Semantics:
2165 * Moves the root file system of the current process to the directory put_old,
2166 * makes new_root as the new root file system of the current process, and sets
2167 * root/cwd of all processes which had them on the current root to new_root.
2169 * Restrictions:
2170 * The new_root and put_old must be directories, and must not be on the
2171 * same file system as the current process root. The put_old must be
2172 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2173 * pointed to by put_old must yield the same directory as new_root. No other
2174 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2176 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2177 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2178 * in this situation.
2180 * Notes:
2181 * - we don't move root/cwd if they are not at the root (reason: if something
2182 * cared enough to change them, it's probably wrong to force them elsewhere)
2183 * - it's okay to pick a root that isn't the root of a file system, e.g.
2184 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2185 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2186 * first.
2188 asmlinkage long sys_pivot_root(const char __user * new_root,
2189 const char __user * put_old)
2191 struct vfsmount *tmp;
2192 struct nameidata new_nd, old_nd;
2193 struct path parent_path, root_parent, root;
2194 int error;
2196 if (!capable(CAP_SYS_ADMIN))
2197 return -EPERM;
2199 error = __user_walk(new_root, LOOKUP_FOLLOW | LOOKUP_DIRECTORY,
2200 &new_nd);
2201 if (error)
2202 goto out0;
2203 error = -EINVAL;
2204 if (!check_mnt(new_nd.path.mnt))
2205 goto out1;
2207 error = __user_walk(put_old, LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old_nd);
2208 if (error)
2209 goto out1;
2211 error = security_sb_pivotroot(&old_nd.path, &new_nd.path);
2212 if (error) {
2213 path_put(&old_nd.path);
2214 goto out1;
2217 read_lock(&current->fs->lock);
2218 root = current->fs->root;
2219 path_get(&current->fs->root);
2220 read_unlock(&current->fs->lock);
2221 down_write(&namespace_sem);
2222 mutex_lock(&old_nd.path.dentry->d_inode->i_mutex);
2223 error = -EINVAL;
2224 if (IS_MNT_SHARED(old_nd.path.mnt) ||
2225 IS_MNT_SHARED(new_nd.path.mnt->mnt_parent) ||
2226 IS_MNT_SHARED(root.mnt->mnt_parent))
2227 goto out2;
2228 if (!check_mnt(root.mnt))
2229 goto out2;
2230 error = -ENOENT;
2231 if (IS_DEADDIR(new_nd.path.dentry->d_inode))
2232 goto out2;
2233 if (d_unhashed(new_nd.path.dentry) && !IS_ROOT(new_nd.path.dentry))
2234 goto out2;
2235 if (d_unhashed(old_nd.path.dentry) && !IS_ROOT(old_nd.path.dentry))
2236 goto out2;
2237 error = -EBUSY;
2238 if (new_nd.path.mnt == root.mnt ||
2239 old_nd.path.mnt == root.mnt)
2240 goto out2; /* loop, on the same file system */
2241 error = -EINVAL;
2242 if (root.mnt->mnt_root != root.dentry)
2243 goto out2; /* not a mountpoint */
2244 if (root.mnt->mnt_parent == root.mnt)
2245 goto out2; /* not attached */
2246 if (new_nd.path.mnt->mnt_root != new_nd.path.dentry)
2247 goto out2; /* not a mountpoint */
2248 if (new_nd.path.mnt->mnt_parent == new_nd.path.mnt)
2249 goto out2; /* not attached */
2250 /* make sure we can reach put_old from new_root */
2251 tmp = old_nd.path.mnt;
2252 spin_lock(&vfsmount_lock);
2253 if (tmp != new_nd.path.mnt) {
2254 for (;;) {
2255 if (tmp->mnt_parent == tmp)
2256 goto out3; /* already mounted on put_old */
2257 if (tmp->mnt_parent == new_nd.path.mnt)
2258 break;
2259 tmp = tmp->mnt_parent;
2261 if (!is_subdir(tmp->mnt_mountpoint, new_nd.path.dentry))
2262 goto out3;
2263 } else if (!is_subdir(old_nd.path.dentry, new_nd.path.dentry))
2264 goto out3;
2265 detach_mnt(new_nd.path.mnt, &parent_path);
2266 detach_mnt(root.mnt, &root_parent);
2267 /* mount old root on put_old */
2268 attach_mnt(root.mnt, &old_nd.path);
2269 /* mount new_root on / */
2270 attach_mnt(new_nd.path.mnt, &root_parent);
2271 touch_mnt_namespace(current->nsproxy->mnt_ns);
2272 spin_unlock(&vfsmount_lock);
2273 chroot_fs_refs(&root, &new_nd.path);
2274 security_sb_post_pivotroot(&root, &new_nd.path);
2275 error = 0;
2276 path_put(&root_parent);
2277 path_put(&parent_path);
2278 out2:
2279 mutex_unlock(&old_nd.path.dentry->d_inode->i_mutex);
2280 up_write(&namespace_sem);
2281 path_put(&root);
2282 path_put(&old_nd.path);
2283 out1:
2284 path_put(&new_nd.path);
2285 out0:
2286 return error;
2287 out3:
2288 spin_unlock(&vfsmount_lock);
2289 goto out2;
2292 static void __init init_mount_tree(void)
2294 struct vfsmount *mnt;
2295 struct mnt_namespace *ns;
2296 struct path root;
2298 mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
2299 if (IS_ERR(mnt))
2300 panic("Can't create rootfs");
2301 ns = kmalloc(sizeof(*ns), GFP_KERNEL);
2302 if (!ns)
2303 panic("Can't allocate initial namespace");
2304 atomic_set(&ns->count, 1);
2305 INIT_LIST_HEAD(&ns->list);
2306 init_waitqueue_head(&ns->poll);
2307 ns->event = 0;
2308 list_add(&mnt->mnt_list, &ns->list);
2309 ns->root = mnt;
2310 mnt->mnt_ns = ns;
2312 init_task.nsproxy->mnt_ns = ns;
2313 get_mnt_ns(ns);
2315 root.mnt = ns->root;
2316 root.dentry = ns->root->mnt_root;
2318 set_fs_pwd(current->fs, &root);
2319 set_fs_root(current->fs, &root);
2322 void __init mnt_init(void)
2324 unsigned u;
2325 int err;
2327 init_rwsem(&namespace_sem);
2329 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount),
2330 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
2332 mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
2334 if (!mount_hashtable)
2335 panic("Failed to allocate mount hash table\n");
2337 printk("Mount-cache hash table entries: %lu\n", HASH_SIZE);
2339 for (u = 0; u < HASH_SIZE; u++)
2340 INIT_LIST_HEAD(&mount_hashtable[u]);
2342 err = sysfs_init();
2343 if (err)
2344 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
2345 __FUNCTION__, err);
2346 fs_kobj = kobject_create_and_add("fs", NULL);
2347 if (!fs_kobj)
2348 printk(KERN_WARNING "%s: kobj create error\n", __FUNCTION__);
2349 init_rootfs();
2350 init_mount_tree();
2353 void __put_mnt_ns(struct mnt_namespace *ns)
2355 struct vfsmount *root = ns->root;
2356 LIST_HEAD(umount_list);
2357 ns->root = NULL;
2358 spin_unlock(&vfsmount_lock);
2359 down_write(&namespace_sem);
2360 spin_lock(&vfsmount_lock);
2361 umount_tree(root, 0, &umount_list);
2362 spin_unlock(&vfsmount_lock);
2363 up_write(&namespace_sem);
2364 release_mounts(&umount_list);
2365 kfree(ns);